OLE

Note:
Only very basic description of the project is presented here - for detailed information, please refer to the recent materials offered in the publications section.

System's purpose

• OLE provides a modular platform for incorporation and cross-application of various existing or OLE-specific knowledge extraction methods.
• It supports universal integration of NLP preprocessing tools - mainly taggers, tokenisers and chunk-parsers for English and possibly various other languages.
• We have developed a framework for innate uncertainty representation and so called empirical conceptual refinement. This serves also as a groundwork for design of novel methods and algorithms of empirical uncertain inference.
• This is directly related to yet another perspective of OLE as an implementation basis for expressive, robust, efficient and Open World Assumption aware reasoning services.

Architecture of the OLE core

Description of the OLE core components

• Resources: relevant documents in plain text, PDF, PostScript or HTML provided by external tools (document classifiers, existing databases of related resources etc.).
• C. L. U. S. Tools: cross-language universality support tools that allow OLE to preprocess data in a given language under specific conditions (namely existence of POS tagged corpora for training of OLE's internal tagger).
• OLITE: the core of the extraction module responsible for creation of ontologies according to provided preprocessed data, utilising some of the knowledge extraction plugins.
• KE Plugins: knowledge extraction plugins implementing various methods for miniontology (an ontology representing one input document) or even whole domain ontology generation.
• Add. and Conf. Files: additional and configuration files (such as a file containing explicit patterns for pattern-based extraction method).
• M. O. Dump: direct dump of miniontology as a product of OLITE module.
• OLEMAN: module that merges the miniontologies resulting from the OLITE module and updates the base domain ontology. Covers also the reasoning implementation and query-processing interface.
• D. O.: continuously updated domain ontology.

Current state of the project

• The modular OLE core framework has been implemented - it is able to incorporate various existing knowledge extraction methods (pattern-based, clustering techniques, formal concept analysis etc.); the framework contains also needful NLP modules and tools supporting possible cross-language universality.
• A novel ANUIC framework for representation of fuzzy and empirical knowledge has been designed and implemented.
• So called conceptual refinement mechanism has been devised, based on the representation framework - it conforms to the general human-inspired and Open World Assumption motivations and allows to boost the precision of primary methods of knowledge acquisition under a certain perspective.
• Pattern-based method and newly devised and implemented clustering and annotation technique have been applied, combined with the conceptual refinement, in the task of taxonomy acquisition.
• A computer-science taxonomy with 1,584 classes, 30,815 individuals and 1,293,998 mutual imprecise is-a relations has been produced from resources containing 20,405,014 words in 3,272 documents; the precision of initial methods was boosted by approximately 35%, when considering only the most relevant relations in the resulting ontology.
• SOLE web interface for automatic custom ontology generation from provided textual resources has been implemented recently; it is based on application of user-defined patterns for arbitrary relations and the OLE core clustering and annotation technique.

Future work

• invention, implementation and verification of formal calculus for empirical and possibly distributed automatic inference among fuzzy ontologies - this consists roughly of these subtasks:
  1. logical underpinning - development of formal (model-theoretic) semantics based on fuzzy relations and fuzzy sets induced by the edges of the proposed ANUIC format;
  2. utilisation of the graph-based fuzzy structure - devision of rigorous mechanisms of incorporation and interpretation of additional general or domain-specific fuzzy rules and relations;
  3. consequent implementation of the reasoning engine - this is motivated by the need for various general inference services like: computation of subsumption hierarchy, instance-checking, identification of least common subsumer, identification of most common concept, concept definition unification and concept definition rewriting;
  4. besides the services mentioned in the previous point, the reasoner should be effective also in the more common tasks: inconsistency handling, query processing, versioning, version tracking and change detection, ontology evolution, common-sense reasoning and many other...
• investigation of methods of bidirectional conversion between the proposed format and common standard languages like (fuzzy) DLs - using mainly alpha-cut abstractions applied on the fuzzy network structure and syntactic transformation or restriction rules; besides the inter-operability dimension, this should be promising for example for relevance measurement in large classical (inconsistent) knowledge bases (using composition of both directions of the mutual transformation procedure);
• development of vast and complex empirical ontologies - using our current tools supported by incorporation of modules covering existing methods of extensive knowledge extraction or implementation of interface to the existing tools (like GATE);
• devision of specific reasoning rules and interpretation perspectives for the developed knowledge repositories, their exploitation by the empirical inference engine;
• development and application of methods of extensive formal evaluation of automatically gained complex ontologies and the implemented reasoning techniques;
• continuous update of the implementation of SOLE interface to OLE core, extending its current functionalities according to the progressive research; corresponding API implementation.